Article: Social Identity in Communities
Reviewed: November 2026 · Method version 2 · How these reviews work
Each factual claim in the article was checked against the source it cites, then challenged by reviewers whose job was to find the strongest evidence against it: replication attempts, meta-analyses, later papers that cite the original work, and research from other fields that study the same mechanism. A separate reviewer weighed both sides, and a script re-checked every quoted source. Where a claim was overstated or misattributed, the article was corrected. The research below supports the article as it now reads.
In a follow-up look at combinations of features, members who could see their group’s page and kept seeing their group’s recent activity visited almost twice as often as everyone else: 11.6 times on average, against 5.7.
The 11.6 versus 5.7 comparison comes from the authors’ post hoc analysis of feature combinations; the planned comparison found 44 percent more visits for the identity features.
Getting the most money for everyone barely moved their choices. Getting more for their own group did. What moved them most was the size of the gap between their group and the other one, which they pushed for “even at the price of sacrificing other ‘objective’ advantages.”
Tajfel et al. also found that maximizing their own group’s profit shaped choices. A later meta-analysis of cooperation studies found that group bias mostly takes the form of favoring one’s own side rather than harming the other.
Work on the social identity model of deindividuation effects (SIDE) suggests that when people online share a social identity, anonymity can make them more open to group influence and more prone to stereotyping and discrimination.
Later SIDE experiments found the effect depends on a clear group norm and, in one study, held only for women.
Social psychologists would not have been surprised. For half a century, experiments have shown that being sorted into a group by a label alone is enough to make people favor it.
Minimal group experiments show that a label alone produces in-group favoritism in how people share rewards; a meta-analysis puts that effect at small without a shared stake (d = 0.19).
Most community products default to bonds: follow, friend, message. The MovieLens experiment tested the two side by side, though members in the bond version looked at its features far less often, which the authors say may explain part of the gap.
In the field experiment, bond-condition members were half as likely to view profiles; in the authors’ follow-up lab study, where exposure was equal, identity and bond features raised attachment about equally. The authors also caution the results may not carry over to friendship-based networks.
In a follow-up survey answered by about one in seven participants, members of the identity condition reported 76 percent more attachment to their movie group.
The 76 percent figure comes from a post-study survey with a 14.2 percent response rate; respondents were more active than nonrespondents.
Scoring that rewards widening the gap over the other side. In Tajfel’s experiments, people gave up rewards to keep that gap wide. Competition between groups can lift effort within each one, so point it at contributions, not at the margin of victory.
A 2026 meta-analysis of 70 effect sizes found that competition between groups raises cooperation within them (g = .677). The caution here is about scoring the gap itself, which Tajfel found people will pay to widen.
These edits were made on November 12, 2026 after the review.
Claim reviewed: In a six-month field experiment, members were sorted into one of ten groups named after animals (Tiger, Eagle, Polar Bear), and each group got a page with its name, an icon, a line about the films it liked, and charts comparing its activity with the other nine.
Now reads: So they ran a six-month field experiment. Every participant was sorted into one of ten groups named after animals: Tiger, Eagle, Polar Bear. In one version of the site, each group got a page with its name, an icon, a line about the films it liked, and charts comparing its activity with the other nine.
Why: The animal names, the ten groups and the comparison charts all match the paper. The one slip is ‘some members were sorted’: every participant got a group, and only the identity versions of the site showed the group pages.
Claim reviewed: Members who could see their group’s page and kept seeing their group’s recent activity visited almost twice as often as everyone else: 11.6 times on average, against 5.7.
Now reads: In a follow-up look at combinations of features, members who could see their group’s page and kept seeing their group’s recent activity visited almost twice as often as everyone else: 11.6 times on average, against 5.7.
Why: The 11.6 versus 5.7 figures are exact. But the paper reports them from a post hoc look at feature combinations, picked out after the main analysis, and the article leads with them as the headline result. The main, planned result was smaller (44 percent more visits).
Claim reviewed: In Tajfel’s closer tests, getting the most money for everyone barely moved choices; what moved them most was the size of the gap between their group and the other one, which they pushed for ‘even at the price of sacrificing other objective advantages.’
Now reads: Getting the most money for everyone barely moved their choices. Getting more for their own group did. What moved them most was the size of the gap between their group and the other one, which they pushed for “even at the price of sacrificing other ‘objective’ advantages.”
Why: The abstract supports ‘barely moved’ for joint profit and ‘moved them most’ for the gap. It also says that getting more for their own group did shape choices, which the article leaves out, so the passage reads as if people only cared about beating the other side.
Claim reviewed: Social identity theory (Tajfel and Turner) starts from the point that groups do not need conflicting interests to compete with or discriminate against each other; being sorted can be sufficient.
Now reads: One of its starting points is that groups do not always need conflicting interests to compete with or discriminate against each other. Being sorted can be sufficient.
Why: The chapter says conflicting interests are ‘not always necessary’ for competition and discrimination; the article’s ‘do not need’ is a touch stronger. A meta-analysis finds categorization alone produces only a slight bias, which fits the article’s hedged ‘can be sufficient’.
Claim reviewed: Social psychologists have known for half a century that a name, a few shared tastes and a rival are enough to create a group people belong to.
Now reads: Social psychologists would not have been surprised. For half a century, experiments have shown that being sorted into a group by a label alone is enough to make people favor it.
Why: The core idea, that an arbitrary label is enough to create a group people favor, is well established and is the article’s own source, though this sentence carries no citation. Two parts stretch it: Tajfel’s groups had no rival, and his studies measured how people split rewards, not belonging or return visits. Without a shared stake, the effect of a label alone is small. Now cites Tajfel, Billig, Bundy & Flament, Social Categorization and Intergroup Behaviour, European Journal of Social Psychology (1971).
Claim reviewed: Usernames and avatars do not dissolve the group; often they leave it as the clearest thing on the screen.
Now reads: The same work carries a design caution. The more a screen shows each person as an individual, with names and personal photos, the weaker the pull of the group tends to be. In one study, groups that saw only group labels grew more attached to their group than groups that saw each other’s first names and photos. Avatars cut both ways: when the group was already in view, members shown with matching avatars fell in with the group more than members shown with different ones.
Why: This sentence has no source, and the research it leans on says the opposite. Postmes, Spears and Lea (2002), now opened directly, found that when people could not see who each other were, group memberships became more salient and individual differences less salient; the groups that saw each other’s first names and photos were the ones where the group faded. Five more experiments in the same tradition agree: names, faces and video make people individuals rather than group members. The one real nuance is about avatars: avatars built as a group uniform strengthen the group, while avatars that make each member look different do not. The article’s ‘often’ has no evidence behind it either way. Now cites Postmes, Spears & Lea, Intergroup differentiation in computer-mediated communication: Effects of depersonalization, Group Dynamics: Theory, Research, and Practice 6(1), 2002; for the avatar sentence, Lee, Effects of Visual Representation on Social Influence in Computer-Mediated Communication, Human Communication Research 30(2), 2004 (https://doi.org/10.1111/j.1468-2958.2004.tb00732.x).
Claim reviewed: The 4,818 participants were randomly assigned to a control, identity or bond version of the site; the MovieLens experiment tested identity and bond head to head.
Now reads: Most community products default to bonds: follow, friend, message. The MovieLens experiment tested the two side by side, though members in the bond version looked at its features far less often, which the authors say may explain part of the gap.
Why: The numbers and random assignment are right. But ‘head to head’ oversells the comparison: members in the bond version looked at its profiles half as often, the authors say the weak bond results may reflect that, and in their follow-up lab study with equal exposure both kinds of features raised attachment about equally.
Claim reviewed: The bond version mirrored each feature at the level of individuals: personal profiles, films the viewer and profile owner both rated highly, and recent activity from people the member had come across before.
Now reads: The bond version mirrored each feature at the level of individuals: personal profiles, films the viewer and the profile owner both rated highly, and recent activity from people the member had come across before or, failing that, people with similar taste.
Why: Profiles and co-rated films match. The feed was not only people the member had met before: when there were too few, it filled in with strangers who had similar taste.
Claim reviewed: Members of the identity condition reported 76 percent more attachment to their movie group.
Now reads: In a follow-up survey answered by about one in seven participants, members of the identity condition reported 76 percent more attachment to their movie group.
Why: The 76 percent figure is in the paper, but it comes from a follow-up survey that only 14 percent of invitees answered, and those who answered were more active than those who did not. It is a self-report from an engaged minority, which the article does not say.
Claim reviewed: Identity features lifted newcomers’ logins by 56 percent and bond features by 20 percent; among old-timers, identity features added 10 percent while bond features cut visits by 16 percent; people with no friends on a site can still belong to something on day one.
Now reads: The split by experience is the part to take into a planning meeting. Among members who had joined in the past month, identity features lifted logins by 56 percent and bond features by 20 percent. Among old-timers, identity features added 10 percent, while bond features cut visits by 16 percent. People with no friends on a site yet can still belong to something early on.
Why: All four subgroup numbers are exact. Two points are stretched: the paper’s ‘newcomers’ had been on the site up to 30 days, not one day, and ‘belong to something on day one’ is the article’s own inference rather than something the study measured.
Claim reviewed: Only 2 percent of participants ever used the new communication tools, so conversation was not the mechanism.
Now reads: Two results keep the case honest. Only 2 percent of participants ever used the new communication tools, so the identity effect did not come from members talking to each other, though the unused tools likely held back the bond version.
Why: The 2 percent figure is exact. ‘So conversation was not the mechanism’ is the article’s inference; the paper simply dropped the tools from its analysis, and notes the failed tools hurt the bond version more than the identity one. The inference is reasonable for the identity effect, but the bond version may have been handicapped by it.
Claim reviewed: Scoring that rewards widening the gap over the other side taps the pull Tajfel measured, and it can cost everyone something.
Now reads: Scoring that rewards widening the gap over the other side. In Tajfel’s experiments, people gave up rewards to keep that gap wide. Competition between groups can lift effort within each one, so point it at contributions, not at the margin of victory.
Why: Tajfel’s result backs the warning that people will give up gains to widen a gap, and field data from Threadless show strong members sabotaging close rivals. But a 2026 meta-analysis finds that competition between groups reliably raises cooperation within them, and the lab finding involved points on paper, not community outcomes, so ‘it can cost everyone something’ is a reasonable caution rather than a demonstrated result.